KR100701163B1 - Methods for Tagging Person Identification Information to Digital Data and Recommending Additional Tag by Using Decision Fusion - Google Patents

Abstract

The present invention relates to a method for extracting with higher accuracy when extracting person identification information from digital data using decision fusion and recommending additional tags other than the existing tags. More specifically, in the digital data management system, when the person identification information is extracted from the digital data according to the attributes and contents of the digital data and automatically assigns the person tag to the digital data, various additional information is provided. Acquiring more accurate person identification information by using and tagging it into digital data, and identifying the tag frequently used in existing digital data assigned the same tag as the tag given to newly created digital data It relates to a method to give in addition to.

Digital Data, Auto Tagging, Face Recognition, Decision Fusion, Bayesian Analysis

Description

Method for Tagging Person Identification Information to Digital Data and Recommending Additional Tag by Using Decision Fusion}

FIG. 1 is a view illustrating a person having a high probability of appearing in a photo the same person as a specific person in a picture when a specific person appears in any one of a plurality of pictures in the same cluster according to a preferred embodiment of the present invention. It is a figure which shows that accuracy can be improved.

Figure 2, according to another preferred embodiment of the present invention, before the two or more people have a meeting to communicate using e-mail, phone calls, SMS messages, etc., the person shown in the picture taken during the meeting is a communication The figure shows that the accuracy of person recognition can be improved by paying attention to the high probability of being the sender or the receiver of the city.

Figures 3a and 3b, according to another preferred embodiment of the present invention, can improve the accuracy of the person recognition by focusing on the fact that only some familiar people among all the people who have been shown in the plurality of pictures often appear in the picture Figure showing that there is.

4 is a diagram illustrating a list of top n lists by combining respective results of person recognition by Bayesian analysis according to a preferred embodiment of the present invention.

5 is a flow chart of a method for recommending additional tags, in accordance with a preferred embodiment of the present invention.

6 is a diagram illustrating a prior art for improving the accuracy of face recognition.

The present invention relates to a method for extracting with higher accuracy when extracting person identification information from digital data and a method for recommending additional tags.

More specifically, in the digital data management system, when the person identification information is extracted from the digital data according to the attributes and contents of the digital data and automatically assigns the person tag to the digital data, various additional information may be applied. To obtain more accurate person identification information and tag it into digital data, and to identify tags that are frequently used in a plurality of existing digital data that have the same tag as that given to newly created digital data. The present invention relates to a method for additionally giving to generated digital data.

Recently, as digital devices such as digital cameras, camera-attached mobile communication devices, digital camcorders, and MP3 players are widely used, the amount and frequency of users generating and exchanging digital data such as digital photographs and digital videos are greatly increased. The expansion of the generation and exchange of digital data inevitably involves data management problems. However, in the case of digital data, since the amount of information is huge, it may not be easy to manage data (such as searching and extracting information) by classifying and integrating data.

As one conventional technique for digital data management, a technique for performing data management such as classifying or integrating data using tags is widely known. A "tag" can be understood as additional data attached to digital data for quick access or retrieval of the data. Such tags typically consist of a series of letters, numbers, or a combination of letters and numbers.

Types of such tags include tags on space, tags on people, tags on objects, tags on time, and the like. Among these, attempts to extract tags on people from digital data with higher accuracy are known. Since.

FIG. 6 discloses a face appearance feature and a contextual feature for face recognition. As a related paper, FIG. 6 is a paper published in 2003 by Microsoft Research Asia, entitled "Family Album." As described in "Automated Annotation of Human Faces in Family Albums", a context-sensitive feature has been added to a general face recognition algorithm in order to increase face recognition rate for a person in a picture. The context characteristic is that an individual appearing in a plurality of photos taken on the same day or at the same event is likely to be wearing the same clothes, and this context characteristic can be used to distinguish a limited number of individuals in an album. According to the paper, face recognition based on this context characteristic does not apply to two pictures taken on different days that are far apart from each other (according to the above paper, more than two days apart). Subject to photographs.

However, the conventional face recognition technology disclosed in the above paper is quite inferior in accuracy because it performs face recognition only on the basis of the similarity of the clothes of the persons within the same time range, in addition to the similarity of the general face shape characteristics. There is this. For example, even if the same time range is set for one or two days, the same person may often wear different clothes. In particular, in the spring, autumn, and winter, except for the summer, people who were wearing jumpers on a shirt are wearing only a shirt indoors. In addition, when the figure of the person is not the front face, there is a fatal disadvantage that not only the recognition of the shape of the face but also the accuracy of the recognition of the shape of the clothes to be worn is deteriorated.

Meanwhile, in the related art, an additional tag is to be additionally added in addition to a tag (for example, at least one of a tag about a place, a tag about a person, a tag about an object, and a tag about a time) assigned when digital data is newly generated. It was inconvenient when you had to enter it manually one by one. On the other hand, a conventional digital data management system is described using Flickr ^TM ( http://www.flickr.com ) (hereinafter referred to as "Flickr") which is a web-based system. When I tagged "dog" and wanted to add an additional tag (e.g. "Poppie") about this dog, I had to manually enter it manually. Increasingly, it was inevitable to be a big disadvantage.

Accordingly, it is an object of the present invention to provide a method for automatically tagging digital data to person identification information generated by performing face recognition with a significantly higher accuracy than a conventional face recognition method. It is to share with other users.

Furthermore, an additional object of the present invention is to identify and use tags frequently used in existing digital data having the same tag as that already assigned to newly generated digital data using decision fusion. In addition to digital data, the user's input burden can be significantly reduced.

DETAILED DESCRIPTION OF THE INVENTION The following detailed description of the invention will be described with reference to the accompanying drawings, which show, by way of illustration, specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein may be embodied in other embodiments without departing from the spirit and scope of the invention with respect to one embodiment. In addition, it is to be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is defined only by the appended claims, along with the full range of equivalents to which such claims are entitled.

In order to achieve the object of the present invention as described above, and to carry out the characteristic functions of the present invention described below, the characteristic configuration of the present invention is as follows:

According to an aspect of the present invention, a method for assigning identification information of a first person as a tag to first digital data generated by a digital device in a digital data management system, wherein the digital data management system identifies the user of the digital device. Information and life pattern information, wherein the first digital data includes an image of the first person, wherein the first person is identified using at least one of additional information about the first person; And automatically assigning identification information of the first person as a tag to the first digital data, wherein the additional information is different from the first digital data, which is distributed in the same cluster as the first digital data. If the first digital data group consisting of at least one other digital data includes an image of the second person, If it is determined that the first person is likely to be the same person as the second person and the first digital data is generated during two or more meetings, the image of the first person is associated with the meeting. M, when the image of the third person is included in the second digital data group composed of at least one digital data that is different from the first digital data, indicating that the communication party is the sender or the receiver of the communication; At least one of information indicating a probability that the first person is the same person as the third person, a generation time and a location of the first digital data based on the number of times; According to the information about the first person obtained by considering together, the position and size of the face of the first person in the image of the first person Information indicating a probability of identifying the first person as a specific person; when the digital device is a dual camera, the first digital data is generated when a lens of the dual camera is directed toward a user of the digital device. The first person is information that the first person is likely to be a user of the digital device, and the first person is generated when the first digital data is not directed toward the user of the digital device when the digital device is a dual camera. A method is provided that includes at least one of information that a probability of not being a user of a digital device is high.

According to another aspect of the present invention, a plurality of existing digital data is stored in a database included in a digital data management system, and a specific user P or a specific object is stored at a specific time zone T of a day at a specific place S using a digital device. When generating first digital data including at least one of the images of O-at least one of the place S, the person P, the object O, and the time zone T is assigned to the first digital data as a primary tag. A second tag different from the primary tag as additional identification information to the first digital data, wherein the same_tag_digital data having the same tag as the primary tag among the existing digital data is searched for Step, the same as the primary tag of all tags assigned to the same tag tag digital data Calculating each frequency of how often the remaining tags except the tag are assigned to the same_tag_digital data, displaying N tags having the highest frequency among the remaining tags on the screen of the digital device; A user selecting at least one of the N tags displayed on the screen as the secondary tag of the first digital data, and assigning the selected tag as the secondary tag to the first digital data; A method is provided.

[Preferred Embodiments of the Invention]

Additional information for person identification 1

1 is a view showing that the accuracy of the recognition of the person can be improved by paying attention to the high probability that the same person is taken in the picture in the same cluster according to a preferred embodiment of the present invention. Here, the clusters may be classified according to the time at which the digital data such as the digital picture is generated or may be classified according to the place where the digital data is generated.

The basis for introducing the cluster concept as described above can be found in the description of the mass tagging portion of Korean Patent Application No. 2006-14040 filed by the same applicant. Accordingly, a user of each digital device can generate digital data at a time and place desired by the user. However, when the generation time and the place of generation of such digital data are examined, there are many cases where the distribution is very discontinuous. That is, a user of a digital device tends to generate digital data only at a very decent time and place. Accordingly, by identifying a place, a time zone, and the like that are common to various digital data, there may be a mass tagging method that bundles the digital data into one chunk and collectively assigns tags common to all the digital data.

This idea of lump tagging can be applied to extract tags about people with high accuracy. That is, when a plurality of digital data is generated in the same cluster, if one digital data includes an image of a specific person, it is likely that the remaining digital data includes an image of the same person. It is likely that a large amount of digital data is generated by different poses about the same person within a short time period, and a large amount of digital data is generated by traveling around the neighborhood while meeting with the same person in a nearby area. This is because they are more likely to generate data, which is a general trend for people.

Therefore, based on such a cluster concept, in digital data tagging information about persons whose images are included in digital data generated by being concentrated in the same time zone or in the same region, the same cluster is used within the same cluster. Using the fact that the same person is likely to appear repeatedly can greatly increase the accuracy of person recognition.

On the other hand, by analyzing the frequency of appearance of people appearing in the image of the digital data already generated in the same cluster to generate a list of candidate groups that are determined to have been taken most often in the same cluster, and then digital data in the same cluster When newly created, one of the most probable persons among the candidate groups already generated is most likely to be determined as the person appearing in the newly generated digital data image. Accordingly, if the identity of the person included in the newly generated image of the digital data is revealed, it is of course possible to update in real time the probability value assigned to each person included in the candidate group list. This process can be done with reference to decision fusion, for example Bayesian analysis, which will be described below.

Additional Information for Recognizing People 2

Figure 2 is a digital data generated during the meeting because the two or more persons according to another preferred embodiment of the present invention is to communicate through email, phone calls, SMS messages, chat (messenger chat), etc. before having a meeting This figure shows that the person represented in the image can increase the accuracy of the person recognition by paying attention to the high probability that the person corresponds to the sender or the receiver during the communication.

As can be seen in the example of the characters shown in Figure 2, it is generally easy for people to contact you to make a decision about an appointment location, time, etc. before meeting a friend. Such contact may be possible through e-mail, telephone call, SMS message exchange using a mobile phone, and chat using a messenger service. Referring to the text exchange shown in FIG. 2 as an example, there was a text contact between Jiro and Tanaka on what time the appointment was scheduled, and after the text contact, it was found that the pictures taken by Jiro and Tanaka were generated. Can be. Therefore, by using the correlation as much as possible, it is possible to increase the accuracy of recognition of the person appearing in the picture.

This additional information can be a very powerful tool given the general trends of people. In order to meet two or more people who are separated from each other and take pictures, it is a requirement that the people who are spaced apart must meet at a certain place after a certain period of time. This is because communication must be made to establish a place and time between people who are spaced apart. It can be seen that the general tendency of human beings can be used appropriately to keep the accuracy of recognition of a person appearing in an image of digital data fairly high.

After the candidate group list is generated by identifying the sender and the receiver of the communication performed during the predetermined period of time before the digital data is generated as described above, when the digital data is newly generated, the most likely candidate group list already generated is assigned. It is very likely that one of the people is judged as the person appearing in the image of the newly generated digital data. Accordingly, if the identity of the person included in the newly generated digital data image is revealed, it may be possible to update in real time with respect to the probability value assigned to each person included in the candidate group list. When a person is included in the image of the digital data, the identity of the person in the image of the digital data is determined by referring to the candidate group list having the updated probability value. This process can be done with reference to decision fusion, for example Bayesian analysis, described below.

Additional Information for Recognizing People 3

3A and 3B illustrate that only some familiar persons among all persons who have been shown in the plurality of photographs frequently appear in the photographs according to another preferred embodiment of the present invention, thereby improving accuracy of person recognition. It is a figure which shows.

In FIG. 3A, when the total number of individual people in all pictures is 100, the x-axis is a sorting order in descending order of the most frequently appearing people in all pictures, and the y-axis indicates the number of pictures when all pictures are assumed to be 100%. At this time, the trajectory of the coordinate values (X, Y) on the graph shown on the xy plane indicates that the total number of Y pictures of the top X people who appear most frequently in the picture appears.

For example, according to FIG. 3A, it can be seen that the photos of the top nine who appear most frequently in all photographs correspond to 37% of the total photographs. This figure is only 9%, considering that the total number of people is 100, but the percentage of these 9% accounts for up to 37% of the total picture. As shown in FIG. 3A, the graph is very biased to the upper left side. It can be seen.

A graph that is directly proportional to the xy plane is that all the people appear in the photo at the same frequency, so it's not useful at all as additional information for face recognition, but in fact, people tend to take more photos with people they're acquainted with. Therefore, as shown in FIG. 3A, if the graph has a shape that deviates significantly from the direct proportional graph, it has significant utility value as additional information for face recognition, and it is possible to increase the accuracy of recognition of the person in the picture based on the correlation. Will be.

3B is bar graphs (X, Y) when the x-axis represents the number of pictures and the y-axis represents the number of people. Here, the bar graphs (X, Y) indicate that Y people have X pictures of their faces. For example, a Y value of 24 for an X value of 1 means that in all photos, 24 people have one photo of their face, and a Y value of 40 for an X value of 1 means that when we consider all the pictures, one person has as many as 40 pictures of his face.

3B, three dotted box shapes appear. Since the rightmost dotted box is the person who is photographed at the highest frequency, it is more likely to be a self shot or a family picture, and the middle dotted box is photographed at the next higher frequency. This person is most likely a best friend because the person is taken, and the left dotted box is most likely a best friend since the person was taken the next highest frequency. Those who belong to the bar graph on the left side of the left dashed box are more likely to be people who do not have a close relationship with each other. Of course, there may be individual differences. Of course, it can be reflected through. For example, the person in the right-most dotted box may be a best friend rather than a self-shot or family member, which receives feedback about the surrounding relationships of a particular person, ie, portrait analysis of previously taken pictures. By receiving the feedback through, it is possible to modify the identity of the group of people corresponding to each dotted box, as well as to improve the accuracy of face recognition for the people appearing in the picture.

In short, after a candidate list is created that lists the top person who has the highest frequency among all the characters in the digital data already created, and the digital data is newly generated, the probability is most likely among the already created candidate list. It is very likely that one of the figures is judged as the person appearing in the image of the newly generated digital data. Through this process, if a person's identity appears in the newly created digital data image, it is possible to update in real time with respect to the probability value assigned to each person included in the candidate group list. When a person is included in the image of the digital data, the identity of the person in the image of the digital data is determined by referring to the candidate group list having the updated probability value. This process can be done with reference to decision fusion, for example Bayesian analysis, described below.

Additional Information for Recognizing People 4

In order to increase the accuracy of recognition of the face of the person in the picture, a tag automatically or manually tagged in the picture may be referred to. For example, you can refer to spatial tags or time tags that are automatically or manually tagged. As described above, the tag may be referred to and life pattern information of a specific person may be added to recognize the person in the picture.

For example, if a photo was created on the weekend (that is, if the time tag in the digital data falls on the weekend), you can weight the person in the photo most likely to be a family or friend, or the photo was created during the week. Can be weighted if the person in the picture is more likely to be a colleague, and when the picture was taken at the bar (i.e., the spatial tag of the digital data is at the bar). If applicable, you can weight the person in the picture to be more likely to be a friend or colleague, or if the picture was taken at home (i.e. if the spatial tag of the digital data is at home), the person in the picture is a family member. You can also weight the probability.

Similarly, if you have a different lifestyle, for example, a lifestyle where you don't meet at a bar with a friend, usually at an amusement park, or if you are a single student who is not alone with your family or friends on the weekend, Of course, by automatically analyzing the life patterns of the past photos of the weight of the probability can be modified according to the life pattern.

In short, a list of candidates whose names are most likely to be judged as people who appear in images of digital data generated at specific time zones and regions by analyzing and understanding the life patterns or special habits of specific people appearing in the digital data already generated. After the list is generated, when digital data is newly generated, it is highly likely that one of the most probable persons among the candidate groups already generated is judged as the person appearing in the image of the newly created digital data. . Through this process, if a person's identity appears in the newly created digital data image, the real-time update may occur for the probability value assigned to each person included in the candidate group list. When a person is included in the image of the digital data, the identity of the person in the image of the digital data is determined by referring to the candidate group list having the updated probability value. This process can be done with reference to decision fusion, for example Bayesian analysis, described below.

Additional Information for Recognizing People 5

Analyzing the position and size of the face in the picture shows that it can be used to increase the accuracy of face recognition. For example, if one big face appears in the center of the picture, it could be a self-shot of the person. If a big face appears in the center of the picture, and a very small face around it, the big face in the center is a selfie. Faces around them are more likely to become strangers (eg pedestrians). Also, if two faces of the same size appear and one of the two faces is already tagged with myself, the identity of the other face may be my best friend or girlfriend. At this time, as described above, according to the analysis of the people appearing in the past picture, that is, the real-time updater may occur with respect to the weight of the probability, depending on the general tendency or special behavior of the main character of the picture. This process can be done with reference to decision fusion, for example Bayesian analysis, described below.

There are many other ways to improve the accuracy of face recognition. For example, in the case of dual cameras, the identity of the person shot through the lens facing the camera user may be more likely to be himself or herself. The identity of the person shot through the lens facing the camera may be a friend except me. Assuming self-shots make up a significant portion of the entire picture, the accuracy of face recognition increases because the self-shots are excluded if the lens is not facing the camera user.

Decision  Combining Additional Information for Recognizing People through Fusion

4 corresponds to a person included in an image of digitally generated digital data by combining the clues shown in the several preferred embodiments mentioned above through decision fusion, e.g. Bayesian analysis, taking into account all the factors in total. This is a diagram illustrating a process of making a list of the top n people, aka top n lists, with the highest probability of becoming a candidate. n is a natural number, for example, let n = 9.

Decision Fusion is an iterative process for answering scientific questions by combining information that accumulates over time, evaluating the current state of the subject matter and collecting new data to find answers to the remaining questions. It is an analysis method that has the effect of updating knowledge and reducing errors by combining new data collected with old data. In short, it refers to the process of making a decision on a specific issue from multiple cues and a priori probability.

There are several ways of decision fusion, which involves taking the probabilities from each of the clues, then combining those probabilities (e.g., weighted probability sum) to get the final probabilities to find the top n list. Unlike the hoc method and the ad hoc method, there are various methods ranging from Bayesian analysis ( http://en.wikipedia.org/wiki/Bayesian_inference ) to calculate the probability considering each clue and obtain the Top n list. In general, clues are not mutually independent (time and place are often correlated; for example, a salaryman will usually be in the office during the day of the week). As in the ad hoc method, it may not be optimal to obtain a probability from each of the clues, and then combine the probabilities to obtain the final probability, so that the present invention prefers the implementation by Bayesian analysis, but the ad hoc method It is clear that this does not preclude implementation.

Through the Bayesian analysis method, the Top 9 list may be generated by combining each of the additional information (list of candidate groups obtained from the above-mentioned embodiments) for person recognition. As shown in FIG. 4, the content of the box on the left shows the probability of each of the 100 persons calculated using all of the above-mentioned additional information for recognizing the person, assuming that there are 100 people in the picture (represented in the picture). Probability of being judged as a person). Due to the unique nature of Bayesian methods, the probability described in this box is continuously updated whenever new digital data is generated, and based on the updated probability, the identity of the person appearing in the image of digital data to be generated is estimated. Each probability is changed dynamically to allow accurate prediction. This Bayesian analysis generates the Top 9 list as the right box. Referring to the created Top 9 list, it is possible to remove a large part of errors that may occur when face recognition for the person in the picture.

Additional tag recommendation function

5 is a flow chart for a method for recommending additional tags, in accordance with a preferred embodiment of the present invention.

When a plurality of existing digital data are stored in a DB of a digital data management system, a specific user uses a digital device to display digital data including at least one of an image of a specific person or a specific object at a specific time of day in a specific place. Create (step 501). At this time, at least one of a place, a person, an object, and a time zone is assigned as a tag to the generated digital data, and these tags are automatically assigned to the digital data together with the generation of the digital data (the Korean patent application filed by the same applicant). No. 2006-14040).

Thereafter, the digital data having the same tag as the tag automatically assigned to the generated digital data is retrieved from the DB (step 502).

Then, when all of the tags assigned to the one or more digital data for which the search is completed are referred to as the remaining tags except for the same tag in step 502, the frequency of each is calculated how often each of the remaining tags is assigned to the retrieved digital data. (Step 503).

Thereafter, the N tags with the highest frequency among the remaining tags are displayed on the screen of the digital device to give the user a chance to select (step 504), and the user selects at least one of the N tags displayed on the screen ( Step 505).

Accordingly, the selected item is added to the newly generated digital data to function as an additional tag (step 506).

For example, when a particular user takes a picture with her girlfriend, most of the time they are shopping at a particular department store, they often tag the digital data with "shopping" and also tag the digital data with "dating." Assume that there are many times to attach. As can be seen in the specification of Korean Patent Application No. 2006-14040 mentioned above, the digital data generated by a specific user is automatically identified as a 'specific department store' by finding a place where the digital data is generated by tracking the position of the digital device by GPS. The name of the girlfriend can be automatically tagged as a tag if it is found that the person in the picture is a girlfriend by performing face recognition on the person in the picture according to the present invention. When searching for digital data having the same tag in the DB using search terms such as place tag and person tag automatically assigned, the tag that appears most frequently in these digital data is probably called 'shopping' or ' It might be a tag called "date." This frequency can be calculated by decision fusion, e.g. Bayesian analysis, to generate the top n list, e.g., the Top 9 list, of the most frequent n tags, and display them on the screen of the digital device. To allow the user to make choices. Therefore, the user-friendly tag input method that could not be implemented in the prior art flicker, etc. can be implemented.

Meanwhile, the top n list is updated in real time by Bayesian analysis, so when another digital data is generated, the user can select an additional tag from the updated top n list by reflecting the tag of the digital data generated immediately before. It can be provided to.

The present invention enables to extract with higher accuracy when extracting the person identification information from the digital data, specifically, in the digital data management system, by extracting the person identification information from the digital data according to the attributes and content of the digital data When automatically tagging digital data on the basis, the additional information on the people in the picture is utilized so that the person identification information can be tagged more accurately.

In addition, the present invention, by using the Fusion Fusion to identify the tag frequently used in the existing digital data having the same tag as the tag given to the newly created digital data and additionally assign it to the digital data input of the user It is a very relieving burden.

Although the present invention has been described in connection with various embodiments as described above, it is to be understood that other variations and modifications may be implemented without departing from the spirit and scope of the invention as would be understood by those skilled in the art. It should be understood. Such modifications and variations are to be considered as falling within the scope of the invention and the appended claims.

Claims (19)

A method of assigning identification information of a first person as a tag to first digital data generated by a digital device in a digital data management system, wherein the digital data management system includes identification information and life pattern information of a user of the digital device. The first digital data comprises an image of the first person, Identifying the first person using at least one of the additional information about the first person, and Automatically assigning identification information of the first person as a tag to the first digital data; The additional information, (a) the first person when the first digital data group consisting of at least one digital data different from the first digital data distributed in the same cluster as the first digital data includes an image of a second person; Information indicating a high probability of being the same person as the second person, (b) if it is determined that the first digital data was generated during two or more encounters, information indicating that the image of the first person is likely to be a sender or a receiver of the communication associated with the encounter, (c) when the image of the third person is included m times in the second group of digital data composed of at least one digital data different from the first digital data, the number of times of the m is increased. Information indicating that there is a high probability of being the same person as 3 persons, wherein m is a natural number; (d) information about the first person obtained by considering at least one of a generation time and a place of the first digital data together with the life pattern information of the user of the digital device; (e) information indicating a probability of identifying the first person as a specific person according to the position and size of the face of the first person in the image of the first person, (f) if the digital device is a dual camera, the first person is a user of the digital device if the first digital data is generated when the lens of the dual camera is directed toward the user of the digital device. Information is high, and (g) if the digital device is a dual camera, the first person is likely to be not the user of the digital device if the first digital data is generated when it is not directed toward the user of the digital device; Containing at least one of Way. The method of claim 1, Identifying the first person comprises combining the additional information by decision fusion, The decision fusion is a generic term for the process of making a decision on a particular issue from multiple cues and a priori probability. Way. The method of claim 2, The decision fusion may include a Bayesian assay or an ad hoc assay. Way. The method of claim 3, The identifying of the first person may include generating a Top n list that is a list of n candidate groups having the highest probability of being the first person based on a result of combining the additional information by the Bayesian analysis or the ad hoc analysis. Containing more Way. The method of claim 4, wherein The Top n list is updated in real time by the Bayesian method or the ad hoc method. Way. The method according to claim 4 or 5, The step of automatically assigning the identification information of the first person as a tag to the first digital data includes: information on the person having the highest probability of being the first person among the candidate groups of the Top n list of the first person; Determining as identification information Way. The communication system of claim 1, wherein the communication is at least one of email communication, wired or wireless telephone call communication, SMS message communication, and messenger communication. Way. The method of claim 1, The (b) information—information about the recipient or originator of the communication associated with the encounter—is obtained from a communication device used in connection with the encounter. Way. The method of claim 1, The (b) information-information about the recipient or originator of the communication associated with the encounter-is obtained by recognizing the content of the communication. Way. The method of claim 1, The (d) information-information about the first person obtained by considering at least one of a generation time and a place of the first digital data together with the life pattern information of the user of the digital device; When the generation time of the first digital data corresponds to a holiday, it is information indicating that the probability that the first person is a family or friend of the user of the digital device is high. When the generation time of the first digital data does not correspond to a holiday, the information indicates that the first person is more likely to be a coworker of the user of the digital device. Way. The method of claim 1, The (d) information-information about the first person obtained by considering at least one of a generation time and a place of the first digital data together with the life pattern information of the user of the digital device; When the place where the first digital data is generated corresponds to the bar, the first person is information indicating that the probability of the first person being a friend or a workmate of the user of the digital device is high. When the place where the first digital data is generated corresponds to a home, the first person is information indicating that the probability of the first person being a family of the user of the digital device is high. Way. The method of claim 11, The generation place of the first digital data is identified by referring to a tag of the first digital data. Way. The method of claim 1, And when the first digital data further comprises an image of a fourth person, identifying the first person according to the information and the position and size of the face of the first person in the image of the first person. Information indicating the probability of identifying by- If the face of the fourth person in the image of the fourth person is located in the center of the entire image of the first digital data, and the face of the fourth person is significantly larger than the face of the first person, the first person Information that indicates a low probability of identifying as a particular person. Way. The method of claim 1, And when the first digital data further comprises an image of a fifth person, identifying the first person according to the information and the position and size of the face of the first person in the image of the first person. Information indicating the probability of identifying by- If the size of the face of the first person is similar to the size of the face in the image of the fifth person, the first person is information indicating that the probability of being a friend of the fifth person is high. Way. The method of claim 14, The fifth person is identified by referring to a tag of the first digital data. Way. A plurality of existing digital data is stored in a database included in the digital data management system, and a specific user includes at least one of an image of a specific person P or a specific object O at a specific time S of the day at a specific place S using a digital device. Generating first digital data, wherein at least one of the place S, the person P, the object O, and the time zone T is assigned as a primary tag to the first digital data; A method of adding a secondary tag as identification information to the first digital data, Searching for identical_tag_digital data having the same tag as the primary tag among the existing digital data; Calculating the frequency of each frequency of how often the remaining tags except the same tag as the primary tag among the tags assigned to the same tag digital data are assigned to the same tag digital data, Displaying the N tags having the highest frequency among the remaining tags on a screen of the digital device; The user selecting at least one of the N tags displayed on the screen as the secondary tag of the first digital data, and Assigning the selected tag to the first digital data as the secondary tag; Way. The method of claim 16, The frequency is calculated by decision fusion, The decision fusion is a generic term for the process of making a decision on a particular issue from multiple cues and a priori probability. Way. The method of claim 17, The decision fusion includes Bayesian analysis or ad hoc analysis Way. The method of claim 18, The N tags are updated in real time Way.

Images